A surgical navigation system based on multimodal imaging

Abstract

The present invention relates to a surgical navigation system based on multimodal imaging, which includes multimodal data acquisition, image preprocessing, image registration, path planning, intraoperative tracking and error correction modules; the system constructs a structural bias registration model, integrates a reversible neural network and a dynamic depth-separable convolution mechanism, and improves the structure preservation ability and registration robustness of multimodal images; introduces a structural gradient-regulated mutual information loss function and a CIDER bias regularization term to enhance the ability to suppress interference from pseudo factors; in terms of path planning, a rare path perception SAC model is proposed, which combines trajectory clustering with a three-dimensional risk map to achieve rare path avoidance and strategy optimization; the system has good registration accuracy, path robustness and tracking consistency, and is suitable for complex navigation scenarios assisted by multimodal images.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a surgical navigation system based on multimodal imaging. Background Art

[0002] With the development of medical imaging technology, the auxiliary role of multimodal medical images such as CT, MRI and ultrasound in the field of surgical navigation is becoming increasingly stronger. Multimodal image fusion can compensate for the incompleteness of single modality information and provide accurate anatomical structure support for preoperative path planning and real-time intraoperative navigation. However, there are many differences in resolution, grayscale value, imaging mechanism, etc. between different modal images, which leads to problems such as structural distortion, interference from pseudo factors and insufficient fusion consistency in the registration process, making it difficult to meet the needs of high-precision surgical navigation. At present, traditional image registration methods mainly use optimization algorithms based on rigid or non-rigid transformations, but such methods often lack the ability to deeply model structural features and have difficulty in handling complex nonlinear differences between modalities. In addition, although existing neural network registration models have introduced deep feature learning mechanisms, their ability to preserve structure and suppress modal bias is still limited. Especially in the presence of strong artifacts or structural ambiguity, registration errors may still accumulate, affecting the robustness and stability of the navigation system.

[0003] On the other hand, preoperative path planning is a key link in the surgical navigation system; although traditional path planning algorithms such as A* and RRT have certain real-time performance, they lack the ability to deeply model historical path patterns and abnormal behavior patterns in complex surgical anatomical scenarios, making it difficult to effectively identify and avoid rare risk paths; at the same time, existing reinforcement learning-based path planning methods mostly use a general strategy training model, which does not consider the rare behavior patterns and environmental risk factors in the actual navigation trajectory during surgery, resulting in poor interpretability and weak robustness of the output path strategy. Summary of the Invention

[0004] The present invention aims to overcome the problems of insufficient structure preservation ability, significant interference of pseudo factors and poor adaptability of path planning models in the existing technology of multimodal image registration, and proposes a surgical navigation system based on multimodal images to achieve system integration optimization of structure bias registration, rare path perception path planning and intraoperative tracking error correction; its innovation lies in: constructing a structure bias registration model, combining reversible neural network and dynamic deep separable convolution mechanism to enhance the local structure preservation ability of cross-modal images; introducing barrier mutual information loss function regulated by structural gradient and kernel estimation CIDER bias regularization term to improve the registration process's resistance to pseudo factors. Robustness to interference from various factors; in terms of path planning, a rare path perception navigation SAC model is proposed, which clusters and identifies historical navigation trajectories, extracts rare behavior patterns, constructs a three-dimensional risk map and embeds it into the strategy training process to achieve risk avoidance and robust optimization of path planning; the present invention constructs a structure preservation and bias suppression mechanism in image registration, introduces rare behavior modeling and risk control strategies in path planning, and combines it with real-time feedback during surgery to achieve comprehensive performance improvement of the navigation system in image processing accuracy, path robustness and tracking stability, and has good technical integration and engineering adaptability.

[0005] The present invention provides a surgical navigation system based on multimodal imaging, which includes a multimodal data acquisition module, an image preprocessing module, an image registration module, a preoperative path planning module, an intraoperative tracking module, and a navigation feedback and error correction module;

[0006] The multimodal data acquisition module collects CT images, MRI images, and ultrasound images, unifies their coordinate system, synchronizes them in time using ECG gating, and generates original multimodal image data;

[0007] The image preprocessing module normalizes pixel intensity, suppresses noise, and enhances ultrasound images on the original multimodal image data to obtain standardized image voxel data. Noise suppression uses non-local mean and 3D Gaussian filtering methods, and ultrasound image enhancement uses Speckle denoising + contrast enhancement technology.

[0008] Image registration module, which builds a structural bias registration model, uses the structural bias registration model to register the standardized image voxel data, and generates a multimodal fusion registration image; the structural bias registration model includes a preprocessing unit, a modality translation unit, a bias optimization unit, a registration unit, a training unit, and a reversible neural network;

[0009] Preoperative path planning module, establishes the SAC model, optimizes the strategy training of the SAC model, constructs the rare path perception navigation SAC model, processes the multimodal fusion registration image through the rare path perception navigation SAC model, and generates the preoperative navigation path. The rare path perception navigation SAC model includes the SAC model;

[0010] The intraoperative tracking module uses visual SLAM technology to perform intraoperative registration based on the preoperative navigation path, matching the preoperative path with the intraoperative image in real time, providing an augmented reality view, performing real-time navigation display, and obtaining real-time position coordinates and intraoperative images;

[0011] The navigation feedback and error correction module corrects the navigation trajectory through Kalman filtering based on real-time position coordinates and intraoperative images.

[0012] Furthermore, the process of registering the standardized image voxel data using the structural bias registration model to generate a multimodal fusion registration image specifically includes the following steps:

[0013] Step S1: normalize the intensity, unify the grayscale range, unify the resolution, and remove the background of the standardized image voxel data to obtain a standardized image pair;

[0014] Step S2: Input the standardized image pair as the source image and the true target image into the reversible neural network respectively, perform modal translation on the source image through the affine coupling structure to generate a pseudo target image, and use the reversible structure to reversely map the pseudo target image to restore the structural information of the source image, ensuring that the structural information is not lost during the modal translation process; embed a dynamic depthwise separable convolution mechanism as a local attention path in the affine coupling structure, perform sparse connection and dynamic weighting through the driven kernel combination, improve the ability to extract local structural details, and generate an enhanced pseudo image; introduce a barrier mutual information loss function regulated by structural gradient, measure the mutual information between the enhanced pseudo image and the true target image, limit the degree of structural distortion between images, and obtain a structure-preserving image;

[0015] Step S3: Define a pseudo factor set, input the structure-preserving image into the backbone network, generate a prediction result, perform conditional distance correlation measurement on the prediction result and the pseudo factor set, introduce a kernel estimation CIDER regularization term to suppress the interference of pseudo factors, implement bias-constrained optimization under structure guidance, and obtain a bias-constrained prediction image;

[0016] Step S4: Input the bias-constrained predicted image and the true target image into the registration network, learn the non-rigid deformation field, and use the spatial transformer to spatially twist the bias-constrained predicted image to generate an initial registered image;

[0017] Step S5: Combine steps S2 to S4 to construct a joint optimization objective function, train the structural bias registration model, further optimize the geometric consistency and modality fusion consistency of the initial registration image, and generate a multimodal fusion registration image.

[0018] Furthermore, the process of generating a preoperative navigation path through the rare path perception navigation SAC model specifically includes the following steps:

[0019] Step B1: extracting surgical anatomical structure information from the multimodal fusion registration images and constructing a 3D sparse voxel map. The 3D sparse voxel map includes a point set and an edge set, where the point set represents the path sampling points and the edge set represents the feasible navigation connectivity, thus forming a structured 3D navigation map.

[0020] Step B2: Based on the structured 3D navigation graph, the state-action pair sequence of the historical preoperative navigation trajectory is collected. The state-action pair sequence is semantically encoded through the encoder network to obtain a feature embedding set. The feature embedding set is then clustered using the K-means algorithm to generate a discretized label sequence, completing the structured abstraction of the trajectory behavior.

[0021] Step B3: Remove duplicates from the discretized label sequence to obtain structural decision patterns, count the frequency of occurrence of each pattern, and sort by frequency to obtain a rare pattern set; based on the rare pattern set, construct the coverage ratio of rare paths;

[0022] Step B4: Based on the rare pattern set and the coverage ratio of rare paths, a three-dimensional spatial risk map is constructed. The three-dimensional spatial risk map is used as external constraint information and embedded into the training environment of the SAC model to avoid high coverage ratios of rare paths and generate trajectory experience pairs. The three-dimensional spatial risk map includes risk scores, rare path coverage ratios, and heat map mapping.

[0023] Step B5: Input the trajectory experience into the SAC model for policy training. Output the action distribution in the current state through the policy network. Estimate the Q function through the dual Q network structure. Monitor the upper bound of the Q function estimation error through the Q value estimation error formula. Output the path planning strategy to obtain the preoperative navigation path.

[0024] By adopting the above scheme, the beneficial effects achieved by the present invention are as follows:

[0025] The present invention realizes the structure preservation and modality alignment of multimodal images in the registration process by constructing a structural bias registration model, effectively improving the geometric consistency and local structure expression ability of the registered images; combining the reversible neural network with the dynamic deep separable convolution mechanism, the registration model has the dynamic modeling ability of local edge features and nonlinear modal differences; further introducing the barrier mutual information loss function and CIDER bias regularization term regulated by structural gradient, effectively suppressing the interference of pseudo factors on the registration accuracy, solving the problem of unstable registration in structural fuzzy areas of existing methods, and enhancing the registration robustness and generalization ability of this system in multimodal image fusion scenarios.

[0026] In terms of path planning, the present invention proposes a rare path perception navigation SAC model, which for the first time combines the clustering modeling mechanism of historical trajectory behavior with the three-dimensional spatial risk map construction method to identify and constrain rare behavior paths in navigation, thereby achieving effective avoidance and robust optimization of abnormal trajectories in the path strategy; the model improves the stability and rationality of the strategy output by introducing rare path coverage ratio and behavior pattern distribution information as training constraints, solves the problems of unreasonable path selection and poor abnormal trajectory avoidance ability of traditional methods, and significantly improves the path generation quality and strategy credibility of the system in complex three-dimensional structure diagrams.

[0027] In addition, the present invention integrates visual SLAM tracking and Kalman filtering error correction mechanism to achieve dynamic matching and position update between the intraoperative path and the real-time image, effectively enhancing the spatiotemporal consistency and system stability during the navigation process; by introducing state estimation and error correction capabilities in continuous feedback during the operation, the system can achieve real-time adjustment of path drift, positioning error and other problems, solving the common problem of navigation accuracy decreasing over time, improving the anti-interference ability and execution accuracy of the entire system during operation, and providing a solid guarantee for the continuity and reliability of the surgical navigation path. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A schematic diagram of a module of a multimodal imaging-based surgical navigation system provided by the present invention;

[0029] Figure 2 This is the navigation feedback and error correction diagram provided in Example 6.

[0030] Figure 2 Middle: blue line: preoperative navigation path, used as a reference; red dashed line: actual trajectory during surgery, obtained by measurement; green dotted line: Kalman filter-corrected trajectory. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0032] Example 1, according to Figure 1 , the present invention provides a surgical navigation system based on multimodal imaging, the system includes a multimodal data acquisition module, an image preprocessing module, an image registration module, a preoperative path planning module, an intraoperative tracking module and a navigation feedback and error correction module;

[0033] The multimodal data acquisition module collects CT images, MRI images, and ultrasound images, unifies their coordinate system, synchronizes them in time using ECG gating, and generates original multimodal image data;

[0034] The image preprocessing module normalizes pixel intensity, suppresses noise, and enhances ultrasound images on the original multimodal image data to obtain standardized image voxel data. Noise suppression uses non-local mean and 3D Gaussian filtering methods, and ultrasound image enhancement uses Speckle denoising + contrast enhancement technology.

[0035] Image registration module, which builds a structural bias registration model, uses the structural bias registration model to register the standardized image voxel data, and generates a multimodal fusion registration image; the structural bias registration model includes a preprocessing unit, a modality translation unit, a bias optimization unit, a registration unit, a training unit, and a reversible neural network;

[0036] Preoperative path planning module, establishes the SAC model, optimizes the strategy training of the SAC model, constructs the rare path perception navigation SAC model, processes the multimodal fusion registration image through the rare path perception navigation SAC model, and generates the preoperative navigation path. The rare path perception navigation SAC model includes the SAC model;

[0037] The intraoperative tracking module uses visual SLAM technology to perform intraoperative registration based on the preoperative navigation path, matching the preoperative path with the intraoperative image in real time, providing an augmented reality view, performing real-time navigation display, and obtaining real-time position coordinates and intraoperative images;

[0038] The navigation feedback and error correction module corrects the navigation trajectory through Kalman filtering based on real-time position coordinates and intraoperative images.

[0039] Example 2: This example is based on Example 1. In this example, the process of registering the standardized image voxel data using the structural bias registration model to generate a multimodal fusion registration image specifically includes the following steps:

[0040] Step S1: normalize the intensity, unify the grayscale range, unify the resolution, and remove the background of the standardized image voxel data to obtain a standardized image pair;

[0041] Step S2: Input the standardized image pair as the source image and the true target image into the reversible neural network respectively, perform modal translation on the source image through the affine coupling structure to generate a pseudo target image, and use the reversible structure to reversely map the pseudo target image to restore the structural information of the source image, ensuring that the structural information is not lost during the modal translation process; embed a dynamic depthwise separable convolution mechanism as a local attention path in the affine coupling structure, perform sparse connection and dynamic weighting through the driven kernel combination, improve the ability to extract local structural details, and generate an enhanced pseudo image; introduce a barrier mutual information loss function regulated by structural gradient, measure the mutual information between the enhanced pseudo image and the true target image, limit the degree of structural distortion between images, and obtain a structure-preserving image;

[0042] Dynamic Depthwise Separable Convolution Mechanism: To improve the ability to extract structural details during image-modal translation, a dynamic depthwise separable convolution mechanism is embedded in the affine coupling structure of the reversible neural network. This mechanism uses an input-driven kernel generation module to generate candidate convolution kernels and enhances structural details in local regions through sparse connections and dynamic weighting. This improves the ability to express local structural information while maintaining the network's lightweight, ensuring geometric consistency between the translated and untranslated images.

[0043] The formula of barrier mutual information loss function regulated by structural gradient is:

[0044] ;

[0045] in, represents the barrier mutual information loss function regulated by structural gradient, represents the enhanced pseudo image, represents the real target image, represents the barrier threshold, represents the linear rectification function, Represents the normalized mutual information between the pseudo target image and the real target image; represents the weight coefficient of the gradient term, express The gradient map of express The gradient map of represents the L1 norm;

[0046] Step S3: Define a pseudo factor set, input the structure-preserving image into the backbone network, generate a prediction result, perform conditional distance correlation measurement on the prediction result and the pseudo factor set, introduce a kernel estimation CIDER regularization term to suppress the interference of pseudo factors, implement bias-constrained optimization under structure guidance, and obtain a bias-constrained prediction image;

[0047] CIDER stands for Conditional Independent Distance Related Regularization Method;

[0048] The loss function in the kernel estimation CIDER regularization term is defined as follows:

[0049] ;

[0050] in, represents the CIDER bias regularization loss function, Represents the prediction result, represents the pseudo factor set, represents a conditional variable, Indicates that given a condition variable Under the premise of and pseudo factors The conditional distance correlation between represents the regularization weight coefficient, represents the high-order dependency penalty term based on kernel estimation;

[0051] Conditional distance correlation is a method used to measure the degree of statistical dependence between two variables given a third conditional variable. It is a dependency measurement method that is extended by introducing conditional variables based on distance correlation.

[0052] The kernel-estimated CIDER regularization term is a novel bias constraint mechanism. It introduces a kernel-estimated approach for the first time to enhance the modeling of complex pseudo-factor dependencies. By integrating a high-order kernel mapping mechanism, it constructs a structure-aware bias control strategy, significantly improving the robustness and interpretability of multimodal registration to pseudo-factor interference.

[0053] Step S4: Input the bias-constrained predicted image and the true target image into the registration network, learn the non-rigid deformation field, and use the spatial transformer to spatially twist the bias-constrained predicted image to generate an initial registered image;

[0054] Step S5: Combine steps S2 to S4 to construct a joint optimization objective function, train the structural bias registration model, further optimize the geometric consistency and modality fusion consistency of the initial registration image, and generate a multimodal fusion registration image. The formula used is as follows:

[0055] ;

[0056] in, represents the total loss function of the joint optimization objective function, represents the smooth regularization term of the non-rigid deformation field, 、 and Represents the weight coefficient.

[0057] Example 3: This example is based on Example 1. In this example, the process of registering the standardized image voxel data to generate a multimodal fusion registered image specifically includes the following steps:

[0058] Step R1: normalize the intensity, unify the grayscale range, unify the resolution, and remove the background of the standardized image voxel data to obtain a standardized image pair;

[0059] Step R2: The standardized image pairs are input into the registration model as the source image and the target image respectively, the source image is modally converted to generate a pseudo target image, and the structure restoration operation is performed to obtain a structure-enhanced image;

[0060] Step R3: Input the structure-enhanced image into the prediction network, output the predicted image result, and perform constraint processing in combination with the predefined pseudo factor set to generate a constrained predicted image;

[0061] Step R4: The constrained predicted image and the target image are input into the registration network to learn the non-rigid deformation relationship between the images, and the registration operation is performed through the spatial transformation module to generate the initial registered image;

[0062] Step R5: Jointly train the network modules involved in the above steps to optimize the registration effect between images and output a multimodal fusion registered image.

[0063] Example 4: This example is based on Example 2. In this example, the process of generating a preoperative navigation path through a rare path perception navigation SAC model specifically includes the following steps:

[0064] Step B1: extracting surgical anatomical structure information from the multimodal fusion registration images and constructing a 3D sparse voxel map. The 3D sparse voxel map includes a point set and an edge set, where the point set represents the path sampling points and the edge set represents the feasible navigation connectivity, thus forming a structured 3D navigation map.

[0065] Step B2: Based on the structured 3D navigation graph, the state-action pair sequence of the historical preoperative navigation trajectory is collected. The state-action pair sequence is semantically encoded through the encoder network to obtain a feature embedding set. The feature embedding set is then clustered using the K-means algorithm to generate a discretized label sequence, completing the structured abstraction of the trajectory behavior.

[0066] Step B3: Remove duplicates from the discretized label sequence to obtain structural decision patterns, count the frequency of occurrence of each pattern, and sort by frequency to obtain a rare pattern set; based on the rare pattern set, construct the coverage ratio of rare paths;

[0067] Structural decision-making patterns include path topology feature patterns, local behavior decision-making patterns, state-action nested sequence patterns, navigation target approach patterns, and operation avoidance strategy patterns;

[0068] These patterns are obtained through: state-action pair sequence → semantic embedding → feature vector clustering → label sequence → deduplication and frequency statistics, resulting in structured trajectory abstract labels. Those with low frequencies are judged as rare path behavior patterns;

[0069] Step B4: Based on the rare pattern set and the coverage ratio of rare paths, a three-dimensional spatial risk map is constructed. The three-dimensional spatial risk map is used as external constraint information and embedded into the training environment of the SAC model to avoid high coverage ratios of rare paths and generate trajectory experience pairs. The three-dimensional spatial risk map includes risk scores, rare path coverage ratios, and heat map mapping.

[0070] Step B5: Input the trajectory experience into the SAC model for policy training. The action distribution in the current state is output through the policy network. The Q function is estimated through the dual Q network structure. The upper bound of the Q function estimation error is monitored through the Q value estimation error formula. The path planning strategy is output to obtain the preoperative navigation path. The formula used is as follows:

[0071] Q value estimation error formula:

[0072] ;

[0073] in, Indicates status, Indicates action, Indicates the strategy The state-action distribution under All Right expectations, represents the optimal Q value, Representation Strategy The Q value under represents the estimation error, represents the discount factor, represents the maximum reward, represents the coverage ratio of rare paths, Represents the approximation error.

[0074] Example 5: This example is based on Example 2. In this example, the process of generating a preoperative navigation path through a rare path perception navigation SAC model specifically includes the following steps:

[0075] Step E1: extracting surgical anatomical structure information from the multimodal fusion registration image and constructing a 3D sparse voxel map. The 3D sparse voxel map includes a point set and an edge set, where the point set represents the path sampling points and the edge set represents the feasible navigation connectivity, thereby forming a structured 3D navigation map.

[0076] Step E2: Based on the structured 3D navigation graph, the state-action pair sequence of the historical preoperative navigation trajectory is collected. The state-action pair sequence is semantically encoded through the encoder network to obtain a feature embedding set. The feature embedding set is then clustered using the K-means algorithm to generate a discretized label sequence, completing the structured abstraction of the trajectory behavior.

[0077] Step E3: Remove duplicates from the discretized label sequence to obtain structural decision patterns, count the frequency of occurrence of each pattern, and sort by frequency to obtain a rare pattern set; based on the rare pattern set, construct the coverage ratio of rare paths;

[0078] Step E4: Based on the rare pattern set and the coverage ratio of rare paths, a three-dimensional spatial risk map is constructed. The three-dimensional spatial risk map is used as external constraint information and embedded into the training environment of the SAC model to avoid high coverage ratios of rare paths and generate trajectory experience pairs.

[0079] Step E5: Input the trajectory experience into the SAC model for policy training, output the action distribution in the current state through the policy network, estimate the Q function through the dual Q network structure, output the path planning strategy, and obtain the preoperative navigation path.

[0080] Example 6, according to Figure 2 This embodiment is based on the fifth embodiment. In this embodiment, the preoperative path planning module establishes a SAC model, optimizes the strategy training of the SAC model, constructs a rare path perception navigation SAC model, processes the multimodal fusion registration image through the rare path perception navigation SAC model, and generates a preoperative navigation path;

[0081] Preoperative navigation path:

[0082] Path length: 85.3 mm; navigation obstacle avoidance radius: 4.5 mm; average spacing: 1.02 mm;

[0083] The intraoperative tracking module uses visual SLAM technology to perform intraoperative registration based on the preoperative navigation path, matching the preoperative path with the intraoperative image in real time, providing an augmented reality view, performing real-time navigation display, and obtaining real-time position coordinates and intraoperative images;

[0084] Camera frame rate: 30 fps;

[0085] Delay is controlled at <80ms;

[0086] Real-time location coordinates:

[0087] t = 0 min: real-time coordinates (134.2, 88.3, ​​45.1);

[0088] t = 7 min: real-time coordinates (139.8, 93.2, 50.6);

[0089] t = 15 min: real-time coordinates (188.5, 123.6, 66.3);

[0090] Navigation feedback and error correction module, which corrects navigation trajectory through Kalman filtering based on real-time position coordinates and intraoperative images;

[0091] Generate navigation feedback and error correction maps.

[0092] The present invention and its embodiments are described above. Such description is not restrictive. What is shown in the accompanying drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in this field are inspired by it and do not depart from the purpose of the invention, they can creatively design structural methods and embodiments similar to the technical solution, which should all fall within the scope of protection of the present invention.

Claims

1. A multimodal imaging-based surgical navigation system, comprising an image preprocessing module, wherein the image preprocessing module generates standardized image voxel data; characterized in that: The system also includes an image registration module and a preoperative path planning module; Image registration module, which builds a structural bias registration model, uses the structural bias registration model to register the standardized image voxel data, and generates a multimodal fusion registration image; The structural bias registration model includes a preprocessing unit, a modality translation unit, a bias optimization unit, a registration unit, a training unit, and a reversible neural network; The preoperative path planning module processes the multimodal fusion registration images through the rare path perception navigation SAC model to generate the preoperative navigation path. The rare path perception navigation SAC model includes the SAC model; The preprocessing unit preprocesses the standardized image voxel data to obtain a standardized image pair; The modal translation unit splits the normalized image pair into a source image and a true target image, which are then fed into a reversible neural network. The source image is modally translated using an affine coupling structure. A dynamic depthwise separable convolution mechanism is embedded in the affine coupling structure as a local attention path. Sparse connections and dynamic weighting are performed through driven kernel combinations to generate enhanced pseudo images. A barrier mutual information loss function regulated by structural gradients is introduced to measure the mutual information between the enhanced pseudo image and the true target image, resulting in a structure-preserving image. The bias optimization unit defines a pseudo-factor set, combines it with the structure-preserving image, generates a prediction result, and performs a conditional distance correlation measurement between the prediction result and the pseudo-factor set. The kernel estimation CIDER regularization term is introduced to suppress the interference of pseudo-factors, and a bias-constrained prediction image is obtained. The process of generating a preoperative navigation path through the rare path perception navigation SAC model includes the following steps: Step B1: Form a structured 3D navigation map based on multimodal fusion and registration images; Step B2: Generate a discretized label sequence based on the structured 3D navigation graph; Step B3: Remove duplicates from the discretized label sequence to obtain the structural decision pattern. Count the frequency of occurrence of each pattern in the structural decision pattern and sort by frequency to obtain a rare pattern set. Based on the rare pattern set, construct the coverage ratio of the rare path. Step B4: Based on the rare pattern set and the coverage ratio of rare paths, a three-dimensional spatial risk map is constructed as external constraint information and embedded into the training environment of the SAC model to avoid high coverage ratios of rare paths and generate trajectory experience pairs; Step B5: Input the trajectory experience into the SAC model for strategy training, estimate the Q function through the dual Q network structure, and monitor the upper bound of the Q function estimation error through the Q value estimation error formula, output the path planning strategy, and obtain the preoperative navigation path.

2. The multimodal imaging-based surgical navigation system according to claim 1, characterized in that: The registration unit predicts the image and the true target image according to the bias constraint and generates an initial registered image.

3. The multimodal imaging-based surgical navigation system according to claim 2, characterized in that: The training unit constructs a joint optimization objective function, trains the structural bias registration model, optimizes the initial registration image, and generates a multimodal fusion registration image.

4. The multimodal imaging-based surgical navigation system according to claim 1, characterized in that: The three-dimensional spatial risk map includes risk scores, coverage ratios of rare paths, and heat map mapping.

Images